{
 "schema": "cain.e93.acceptance_matrix.v1",
 "version": "0.2-interim",
 "generated": "2026-11-18",
 "status": "DRAFT_INTERIM",
 "outcome_rule": "Outcome 1 only if A\u2013K and V pass. Outcome 2 if A, B, I, J pass and at least two of C/D/E/F/G pass. Otherwise Outcome 3.",
 "current_outcome": "OUTCOME 3 (governed agentic intelligence infrastructure). Under the pre-registered rule Outcome 2 requires gate A, which cannot pass as written (cyber = E91 environment reused). Owner decision 2026-11-18: keep the rule as written, so Outcome 3 is the maximum for E93.",
 "gates": [
  {
   "gate": "A",
   "requirement": "Three vertical environments, \u226530 held-out each",
   "status": "NOT_PASSED_AS_WRITTEN",
   "evidence": "Software + science built in E93; cyber = E91 environment reused (owner decision). 90 held-out tasks frozen 2068c41e9."
  },
  {
   "gate": "B",
   "requirement": "Solver-independent hidden checkers",
   "status": "PASS",
   "evidence": "Sandboxed checkers; bypass found by S92 and fixed (H1/H2, b00b22ac6) before held-out use; 8 regression tests."
  },
  {
   "gate": "C",
   "requirement": "Learning vs no-learning, Holm-corrected",
   "status": "PASS_INTERNAL",
   "evidence": "Synthesis learning curve k0\u2192k60: 87\u2192369, 114\u2192367 (p<1e-76). Internal families; vertical learning pending. Software vertical: learned repair ranking L_30 16/30 vs F 3/30 (p=0.00235), vs no-learning L_0 6/30 (11 vs 1, p=0.00635, exploratory). In-distribution (targets drawn from the search language); same-team task authorship; automated same-process approver."
  },
  {
   "gate": "D",
   "requirement": "\u22655 new procedures, validated, beating selection",
   "status": "PARTIAL",
   "evidence": "12+12 promoted, 9/10 behaviourally new, beat G significantly; at matched compute equal to selection. In-distribution (targets drawn from the search language); same-team task authorship; automated same-process approver. Model-proposed candidates 0/34 valid."
  },
  {
   "gate": "E",
   "requirement": "\u22652 cross-domain transfer experiments with controls",
   "status": "FAIL_SO_FAR",
   "evidence": "Within-domain cross-family: 0 positive transfers; cross-domain not run."
  },
  {
   "gate": "F",
   "requirement": "Adaptive \u2265 best fixed at matched budget",
   "status": "PASS_INTERNAL",
   "evidence": "Adaptive 390/418 beats G2 at lower cost; vertical version pending. In-distribution (targets drawn from the search language); same-team task authorship; automated same-process approver."
  },
  {
   "gate": "G",
   "requirement": "Heterogeneous team vs best single agent",
   "status": "NOT_RUN",
   "evidence": ""
  },
  {
   "gate": "H",
   "requirement": "\u22656 h mission on vertical tasks, faults, independent verification",
   "status": "IN_PROGRESS",
   "evidence": "Started 12:52:26Z; ends ~19:10Z."
  },
  {
   "gate": "I",
   "requirement": "Governance lab + kill switch measured at resource",
   "status": "PASS_INTERNAL",
   "evidence": "19/19, 10/10, 20/20; same host; no external review."
  },
  {
   "gate": "J",
   "requirement": "Signed bundle + offline verifier from raw rows",
   "status": "PARTIAL",
   "evidence": "Verifier 69/69 (stdlib, no CAIN imports); interim bundle hashed not signed."
  },
  {
   "gate": "K",
   "requirement": "Consistent public disclosure, owner-approved",
   "status": "PARTIAL",
   "evidence": "Interim published on all surfaces by S92 and audited by S66; final pending."
  },
  {
   "gate": "L",
   "requirement": "External review status stated",
   "status": "BLOCKED_STATED",
   "evidence": "No external reviewer; stated on every surface."
  },
  {
   "gate": "V",
   "requirement": "Vertical thresholds \u226570% aggregate, none <60%",
   "status": "NOT_YET_MEASURED",
   "evidence": "Software best arm 16/30 (53%); science and cyber CAIN arms not run. Below the 70% aggregate / 60% per-vertical thresholds so far."
  }
 ]
}